arXiv subjects

Furong Xu

Publications and source records attributed to Furong Xu.

35 records · Page 2Linked to original sources

arXiv2023

Learning Segment Similarity and Alignment in Large-Scale Content Based Video Retrieval

With the explosive growth of web videos in recent years, large-scale Content-Based Video Retrieval (CBVR) becomes increasingly essential in video filtering, recommendation, and copyright protection. Segment-level CBVR (S-CBVR) locates the start and end time of similar segments in finer granularity, which is beneficial for user browsing efficiency and infringement detection especially in long video scenarios. The challenge of S-CBVR task is how to achieve high temporal alignment accuracy with efficient computation and low storage consumption. In this paper, we propose a Segment Similarity and Alignment Network (SSAN) in dealing with the challenge which is firstly trained end-to-end in S-CBVR. SSAN is based on two newly proposed modules in video retrieval: (1) An efficient Self-supervised Keyframe Extraction (SKE) module to reduce redundant frame features, (2) A robust Similarity Pattern Detection (SPD) module for temporal alignment. In comparison with uniform frame extraction, SKE not only saves feature storage and search time, but also introduces comparable accuracy and limited extra computation time. In terms of temporal alignment, SPD localizes similar segments with higher accuracy and efficiency than existing deep learning methods. Furthermore, we jointly train SSAN with SKE and SPD and achieve an end-to-end improvement. Meanwhile, the two key modules SKE and SPD can also be effectively inserted into other video retrieval pipelines and gain considerable performance improvements. Experimental results on public datasets show that SSAN can obtain higher alignment accuracy while saving storage and online query computational cost compared to existing methods.

cs.CV↗

arXiv2023

Boundary-aware Backward-Compatible Representation via Adversarial Learning in Image Retrieval

Image retrieval plays an important role in the Internet world. Usually, the core parts of mainstream visual retrieval systems include an online service of the embedding model and a large-scale vector database. For traditional model upgrades, the old model will not be replaced by the new one until the embeddings of all the images in the database are re-computed by the new model, which takes days or weeks for a large amount of data. Recently, backward-compatible training (BCT) enables the new model to be immediately deployed online by making the new embeddings directly comparable to the old ones. For BCT, improving the compatibility of two models with less negative impact on retrieval performance is the key challenge. In this paper, we introduce AdvBCT, an Adversarial Backward-Compatible Training method with an elastic boundary constraint that takes both compatibility and discrimination into consideration. We first employ adversarial learning to minimize the distribution disparity between embeddings of the new model and the old model. Meanwhile, we add an elastic boundary constraint during training to improve compatibility and discrimination efficiently. Extensive experiments on GLDv2, Revisited Oxford (ROxford), and Revisited Paris (RParis) demonstrate that our method outperforms other BCT methods on both compatibility and discrimination. The implementation of AdvBCT will be publicly available at https://github.com/Ashespt/AdvBCT.

cs.CV↗

arXiv2023

DC-Former: Diverse and Compact Transformer for Person Re-Identification

In person re-identification (re-ID) task, it is still challenging to learn discriminative representation by deep learning, due to limited data. Generally speaking, the model will get better performance when increasing the amount of data. The addition of similar classes strengthens the ability of the classifier to identify similar identities, thereby improving the discrimination of representation. In this paper, we propose a Diverse and Compact Transformer (DC-Former) that can achieve a similar effect by splitting embedding space into multiple diverse and compact subspaces. Compact embedding subspace helps model learn more robust and discriminative embedding to identify similar classes. And the fusion of these diverse embeddings containing more fine-grained information can further improve the effect of re-ID. Specifically, multiple class tokens are used in vision transformer to represent multiple embedding spaces. Then, a self-diverse constraint (SDC) is applied to these spaces to push them away from each other, which makes each embedding space diverse and compact. Further, a dynamic weight controller(DWC) is further designed for balancing the relative importance among them during training. The experimental results of our method are promising, which surpass previous state-of-the-art methods on several commonly used person re-ID benchmarks.

cs.CV↗

arXiv2022

Solutions for Fine-grained and Long-tailed Snake Species Recognition in SnakeCLEF 2022

Automatic snake species recognition is important because it has vast potential to help lower deaths and disabilities caused by snakebites. We introduce our solution in SnakeCLEF 2022 for fine-grained snake species recognition on a heavy long-tailed class distribution. First, a network architecture is designed to extract and fuse features from multiple modalities, i.e. photograph from visual modality and geographic locality information from language modality. Then, logit adjustment based methods are studied to relieve the impact caused by the severe class imbalance. Next, a combination of supervised and self-supervised learning method is proposed to make full use of the dataset, including both labeled training data and unlabeled testing data. Finally, post processing strategies, such as multi-scale and multi-crop test-time-augmentation, location filtering and model ensemble, are employed for better performance. With an ensemble of several different models, a private score 82.65%, ranking the 3rd, is achieved on the final leaderboard.

cs.CV↗

arXiv2022

A Large-scale Comprehensive Dataset and Copy-overlap Aware Evaluation Protocol for Segment-level Video Copy Detection

In this paper, we introduce VCSL (Video Copy Segment Localization), a new comprehensive segment-level annotated video copy dataset. Compared with existing copy detection datasets restricted by either video-level annotation or small-scale, VCSL not only has two orders of magnitude more segment-level labelled data, with 160k realistic video copy pairs containing more than 280k localized copied segment pairs, but also covers a variety of video categories and a wide range of video duration. All the copied segments inside each collected video pair are manually extracted and accompanied by precisely annotated starting and ending timestamps. Alongside the dataset, we also propose a novel evaluation protocol that better measures the prediction accuracy of copy overlapping segments between a video pair and shows improved adaptability in different scenarios. By benchmarking several baseline and state-of-the-art segment-level video copy detection methods with the proposed dataset and evaluation metric, we provide a comprehensive analysis that uncovers the strengths and weaknesses of current approaches, hoping to open up promising directions for future works. The VCSL dataset, metric and benchmark codes are all publicly available at https://github.com/alipay/VCSL.

cs.CV↗

arXiv2022

Nuclear Spectroscopy with Heavy Ion Nucleon Knockout and (p,2p) Reactions

Knockout reactions with heavy ion targets in inverse kinematics, as well as "quasi-free" (p,2p) and (p,pn) reactions are useful tools for nuclear spectroscopy. We report calculations on \textit{ab-initio} many-body wavefunctions based on the no-core shell model to study the nucleon removal reactions in light nuclei, including beryllium, carbon, and oxygen isotopic chains, and explore the importance of using an \textit{ab-initio} method. Our study helps clarifying how the extraction of spectroscopic factors from the experiments depend on the details of the many-body wavefunctions being probed. We show that recent advances with the ab-initio method can provide more insights on the spectroscopy information extracted from experiments.

nucl-th↗

arXiv2021

HBReID: Harder Batch for Re-identification

Triplet loss is a widely adopted loss function in ReID task which pulls the hardest positive pairs close and pushes the hardest negative pairs far away. However, the selected samples are not the hardest globally, but the hardest only in a mini-batch, which will affect the performance. In this report, a hard batch mining method is proposed to mine the hardest samples globally to make triplet harder. More specifically, the most similar classes are selected into a same mini-batch so that the similar classes could be pushed further away. Besides, an adversarial scene removal module composed of a scene classifier and an adversarial loss is used to learn scene invariant feature representations. Experiments are conducted on dataset MSMT17 to prove the effectiveness, and our method surpasses all of the previous methods and sets state-of-the-art result.

cs.CV↗

arXiv2020

From bound states to the continuum

This white paper reports on the discussions of the 2018 Facility for Rare Isotope Beams Theory Alliance (FRIB-TA) topical program "From bound states to the continuum: Connecting bound state calculations with scattering and reaction theory". One of the biggest and most important frontiers in nuclear theory today is to construct better and stronger bridges between bound state calculations and calculations in the continuum, especially scattering and reaction theory, as well as teasing out the influence of the continuum on states near threshold. This is particularly challenging as many-body structure calculations typically use a bound state basis, while reaction calculations more commonly utilize few-body continuum approaches. The many-body bound state and few-body continuum methods use different language and emphasize different properties. To build better foundations for these bridges, we present an overview of several bound state and continuum methods and, where possible, point to current and possible future connections.

nucl-th↗

arXiv2018

Gogny-force derived effective shell-model Hamiltonian

The density-dependent finite-range Gogny force has been used to derive the effective Hamiltonian for the shell-model calculations of nuclei. The density dependence simulates an equivalent three-body force, while the finite range gives a Gaussian distribution of the interaction in the momentum space and hence leads to an automatic smooth decoupling between low-momentum and high-momentum components of the interaction, which is important for finite-space shell-model calculations. Two-body interaction matrix elements, single-particle energies and the core energy of the shell model can be determined by the unified Gogny force. The analytical form of the Gogny force is advantageous to treat cross-shell cases, while it is difficult to determine the cross-shell matrix elements and single-particle energies using an empirical Hamiltonian by fitting experimental data with a large number of matrix elements. In this paper, we have applied the Gogny-force effective shell-model Hamiltonian to the ${\it p}$- and ${\it sd}$-shell nuclei. The results show good agreements with experimental data and other calculations using empirical Hamiltonians. The experimentally-known neutron drip line of oxygen isotopes and the ground states of typical nuclei $^{10}$B and $^{18}$N can be reproduced, in which the role of three-body force is non-negligible. The Gogny-force derived effective Hamiltonian has also been applied to the cross-shell calculations of the ${\it sd}$-${\it pf}$ shell.

nucl-th↗

arXiv2018

Hartree-Fock Gamow basis from realistic nuclear forces

We present a simplified method to generate the Hartree-Fock Gamow basis from realistic nuclear forces. The Hartree-Fock iteration in the harmonic-oscillator basis is first performed, and then the obtained HF potential is analytically continued to the complex-k plane, finally by solving the Schrödinger equation in the complex-k plane the Gamow basis is obtained. As examples, the method is applied to 4He and 22O with the renormalized chiral N3LO potential. The basis obtained which includes bound, resonant and scattering states can be further used in many-body calculations to study weakly bound nuclei.

nucl-th↗

arXiv2016

Isomerism in the "south-east" of $^{132}$Sn and a predicted neutron-decaying isomer in $^{129}$Pd

Excited states in neutron-rich nuclei located south-east of $^{132}$Sn are investigated by shell-model calculations. A new shell-model Hamiltonian is constructed for the present study. The proton-proton and neutron-neutron interactions of the Hamiltonian are obtained through the existing CD-Bonn $G$ matrix results, while the proton-neutron interaction across two major shells is derived from the monopole based universal interaction plus the M3Y spin-orbit force. The present Hamiltonian can reproduce well the experimental data available in this region, including one-neutron separation energies, level energies and the experimental $B(E2)$ values of isomers in $^{134,136,138}$Sn, $^{130}$Cd, and $^{128}$Pd. New isomers are predicted in this region, $e.g.$ in $^{135}$Sn, $^{131}$Cd, $^{129}$Pd, $^{132,134}$In and $^{130}$Ag, in which almost no excited states are known experimentally yet. In the odd-odd $^{132,134}$In and $^{130}$Ag, the predicted very long $E2$ life-times of the low-lying $5^{-}$ states are discussed, demanding more information on the related proton-neutron interaction. The low-lying states of $^{132}$In are discussed in connection with the recently observed $γ$ rays. The predicted $19/2^{-}$ isomer in $^{129}$Pd could decay by both electromagnetic transitions and neutron emission with comparable partial life-times, making it a good candidate for neutron radioactivity, a decay mode which is yet to be discovered.

nucl-th↗

arXiv2016

$\it{Ab}$ $\it{initio}$ nuclear many-body perturbation calculations in the Hartree-Fock basis

Starting from realistic nuclear forces, the chiral N$^3$LO and JISP16, we have applied many-body perturbation theory (MBPT) to the structure of closed-shell nuclei, $^4$He and $^{16}$O. The two-body N$^3$LO interaction is softened by a similarity renormalization group transformation while JISP16 is adopted without renormalization. The MBPT calculations are performed within the Hartree-Fock (HF) bases. The angular momentum coupled scheme is used, which can reduce the computational task. Corrections up to the third order in energy and up to the second order in radius are evaluated. Higher-order corrections in the HF basis are small relative to the leading-order perturbative result. Using the anti-symmetrized Goldstone diagram expansions of the wave function, we directly correct the one-body density for the calculation of the radius, rather than calculate corrections to the occupation propabilities of single-particle orbits as found in other treatments. We compare our results with other methods where available and find good agreement. This supports the conclusion that our methods produce reasonably converged results with these interactions. We also compare our results with experimental data.

nucl-th↗

arXiv2014

Mirror energy difference and the structure of loosely bound proton-rich nuclei around A = 20

The properties of loosely bound proton-rich nuclei around A = 20 are investigated within the framework of nuclear shell model. In these nuclei, the strength of the effective interactions involving the loosely bound proton s1=2 orbit are significantly reduced in comparison with those in their mirror nuclei. We evaluate the reduction of the effective interaction by calculating the monopole-baseduniversal interaction (VMU) in the Woods-Saxon basis. The shell-model Hamiltonian in the sd shell, such as USD, can thus be modified to reproduce the binding energies and energy levels of the weakly bound proton-rich nuclei around A = 20. The effect of the reduction of the effective interaction on the structure and decay properties of these nuclei is also discussed.

nucl-th↗

arXiv2012

Shell evolution in neutron-rich carbon isotopes: Unexpected enhanced role of neutron-neutron correlation

Full shell-model diagonalization has been performed to study the structure of neutron-rich nuclei around $^{20}$C. We investigate in detail the roles played by the different monopole components of the effective interaction in the evolution of the N=14 shell in C, N and O isotopes. It is found that the relevant neutron-neutron monopole terms, $V^{nn}_{d_{5/2}d_{5/2}}$ and $V^{nn}_{s_{1/2}s_{1/2}}$, contribute significantly to the reduction of the N=14 shell gap in C and N isotopes in comparison with that in O isotopes. The origin of this unexpectedly large effect, which is comparable with (sometimes even larger than) that caused by the proton-neutron interaction, is related to the enhanced configuration mixing in those nuclei due to many-body correlations. Such a scheme is also supported by the large B(E2) value in the nucleus $^{20}$C which has been measured recently.

nucl-th↗

arXiv2012

Shell-model study of boron, carbon, nitrogen and oxygen isotopes based on monopole-based-universal interaction

We study boron, carbon, nitrogen and oxygen isotopes with a newly constructed shell-model Hamiltonian developed from monopole-based-universal interaction ($V_{MU}$). The present Hamiltonian can reproduce well the ground-state energies, energy levels, electric quadrupole properties and spin properties of these nuclei in full psd model space including $(0-3)\hbarω$ excitations. Especially, it correctly describes the drip lines of carbon and oxygen isotopes and the spins of the ground states of $^{10}$B and $^{18}$N while some former interactions such as WBP and WBT fail. We point out that the inclusion of $2\hbarω$ excitations is important in reproducing some of these properties. In the present $(0+2)\hbarω$ calculations small but constant E2 effective charges appear to work quite well. As the inclusion of the $2\hbarω$ model space makes rather minor change, this seems to be related to the smallness of $^{4}$He core. Similarly, the spin g factors are very close to free values. The applicability of tensor and spin-orbit forces in free space, which are taken in the present Hamiltonian, is examined in shell model calculations.

nucl-th↗

arXiv2009

Microscopic mechanism of charged-particle radioactivity and generalization of the Geiger-Nuttall law

A linear relation for charged-particle emissions is presented starting from the microscopic mechanism of the radioactive decay. It relates the logarithms of the decay half-lives with two variables, called $χ'$ and $ρ'$, which depend upon the $Q$-values of the outgoing clusters as well as the masses and charges of the nuclei involved in the decay. This relation explains well all known cluster decays. It is found to be a generalization of the Geiger-Nuttall law in $α$ radioactivity and therefore we call it the universal decay law. Predictions on the most likely emissions of various clusters are presented by applying the law over the whole nuclear chart. It is seen that the decays of heavier clusters with non-equal proton and neutron numbers are mostly located in the trans-lead region. The emissions of clusters with equal protons and neutrons, like $^{12}$C and $^{16}$O, are possible in some neutron-deficient nuclei with $Z\geq54$.

nucl-th↗

arXiv2009

Universal decay law in charged-particle emission and exotic cluster radioactivity

A linear universal decay formula is presented starting from the microscopic mechanism of the charged-particle emission. It relates the half-lives of monopole radioactive decays with the $Q$-values of the outgoing particles as well as the masses and charges of the nuclei involved in the decay. This relation is found to be a generalization of the Geiger-Nuttall law in $α$ radioactivity and explains well all known cluster decays. Predictions on the most likely emissions of various clusters are presented.

nucl-th↗